IA continues to be the number one subject of technology, especially in conference rooms and VC circles. By third class 2024, Ey indicated that AI companies represented 37% of all VC investments. But it is not only about money – on the side of employees, 93% believe that AI has a positive impact on their work and even reducing anxiety at work, according to data from The access group.
Throughout the technology sector, AI becomes the engine behind thorny problems such as data management and security and managers are strongly interested in technology to achieve their organizational objectives. But as the AI landscape matures,
Jeff Watkins, CTO at Createfuture, warns that the inflated assessments on the AI market will not last forever. When the bubble bursts, many “chancers” come out, leaving only those who are ready to focus on profitability and tightening their belts. At the same time, some CIOs always await the “Ford moment” with enthusiasm, when automation and productivity really take off.
2025 could finally be the year we are starting to get answers. For the moment, however, here is a quick overview of AI trends that are likely to make waves in the next 12 months.
1. It adopts models of small languages (SLMS)
Steven Webb, head of technology and innovation at Capgemini UK, plans that we are about to see a passage of models of large languages like GPT to smaller and more agile models –Small language models (SLMS). These SLMs are cheaper to execute, need less computing power, waterAnd energyAnd can be adapted to understanding the jargon specific to industry, painful points and technical language. SLMS quickly gain ground with 24% of organizations Already use them, while 54% planned to jump on board in the coming years.
This year has already seen interesting developments in SLMs. Nvidia launched Nim, its own version of SLMS for an Easy Enterprise deployment. At the same time, IBM introduced granite 3.0 SLMS, which could actually surpass the LLM Lama For tasks such as game design, discovery of drugs or code generation due to the use of a limited parameter of 3 to 4 billion. It is much less than the tens or several hundred billions used by the LLM. Salesforce Research is also preparing to deploy SLM that can operate offline, based on data stored on the user device to provide a more personalized and profitable experience.
As pressure around data security, confidentiality and sovereignty Increase, the key to managing these challenges will be to improve how we follow the data line and restricted access only to what is necessary. Currently, it is a limitation of the LLM, but the SLM should fill this shortcoming early.
2. Shadow AI will continue to grow, but AI will also go to block threats
Shadow—OT called “BRING Your OWN IA” – is when employees use unprecedented AI tools in the workplace without the approval of IT or security. Salesforce reports The fact that more than half of generating AI users count on unprecedented tools, and that 7 in 10 workers are not training on how to use these tools in complete safety or ethically. Unsurprisingly, this trend does not slow down.
“The rise of Shadow Ai will create new attack vectors so that the security TEM fights and detect,” explains Nicole Carignan, vice-president of Cyber Ai Strategic in Darktrace. It expects 2025 to be marked by an AI explosion and Generative AI toolsDistributing between businesses and even employee personal devices. Its greatest concern, however, is the financial, reputation and technological consequences that companies could encounter as regulations such as the EU ACT AI Start taking effect.
Shadow AI was at the center of Spotify’s decision to block the use of GPT by employees and to remove “tens of thousands” of tracks created by the boom. According to Financial timeThe movement was aimed at protecting the royalties of artists after suspect streaming models were reported. In the same way, Samsung prohibited GPT After the employees wrongly downloaded the sensitive code to Catrisking initiate attacks.
Nicole maintains that the reduction of these problems requires two things: a continuous food AI asset discovery in real -time training and better management. The way these solutions will shape 2025 is the assumption of anyone. Watkins adds that AI will not only be part of the problem – this will also be part of the solution. He points to a protective AI, which is already used but will make huge steps forward when combined with Agentic in 2025.
The protection AI helps to analyze models in data, network traffic and user behavior to detect threats as phishingmalware, ransomwareand data violations in real time. Once a threat is identified, the protective AI can immediately act – block malware, close the compromise accounts or isolate infected systems. He can also inform the security teams for manual intervention if necessary.
3. AI of agentics will be the start of AI 3.0
Ann Maya, Emea CTO in Boomi, is optimistic about the future of agentic AI, and she has data to support her optimism. “Think of an average expenditure report,” she shares. “An AI financial agent could manage everything, error checks and policies’ compliance with exceptions and approvals management. Everything is possible without human intervention, connecting the departments transparent via APIs linked to sales systems, ERP platformsHR software, and even deeper functions such as entity correspondence in data. »»
AI agents are designed to plan and make decisions by themselves, even when you work with real-time information. Unlike traditional AI systems such as LLMs, which follow “fixed” paths, AI agents can dynamically adjust their actions according to training and context. As Ann explains, with the right guarantees – such as compliance checks and failures – these agents could introduce a more decentralized AI model for companies. For DsiThis means better operational efficiency and unprecedented control over data qualitySafety and inhabitants through smarter and context -oriented decisions.
The agentic AI will not however have a path to easy success. “The key is that companies set out measurable objectives for their AI agents, such as cost reduction or increased efficiency, and to align these objectives with their wider priorities,” explains Ann Itpro. “This is what will make the difference.”
4. AI Modernize inherited infrastructure
Update Inherited technology This often means to face high costs and the risks of making a massive overhaul in one go, which is why it can be difficult for many companies to swallow them. But there is a smarter route – companies will increasingly use AI to perform more manageable, profitable and low risk incremental upgrades.
“AI is no longer a question of” if “, but” when “, explains Burley Kawasaki, global vice-president of strategy at Cretio, when he spoke of the role of AI. He stresses that global companies prioritize inherited modernization, expenses should reach 3.4 dollars by 2026.
According to Burley, companies will start using a combination of IA platforms and without code to update their inherited battery in the coming year. Code -free tools can slowly build and integrate new features with older systems and keep things stable while workflows move to the environment without code. Meanwhile, AI adds intelligence – as predictive analytics and decision -making capacities – to make the process faster and smarter.
5. Multimodal systems will become the new “chatgpt”
The number of AI Multimodal The models increase. Claude can now interpret images in PDFs and documents and chatgpt is now capable of analyzing image prompts, even those integrated into files. These developments indicate a clear thing: multimodal systems are increasing, and by 2025, they will be everywhere.
Multimodal AI brings together different types of data – text, images, audio, video and more – to create a richer understanding of information. Imagine a technician by downloading a photo of an error screen and an AI model able to provide repair guides based on text to solve the problem.
The CIO of Data Intelligence and Nasuni AI, Jim Liddle, sees a multimodal AI becoming a dominant trend in business data management. Its ability to interpret and visualize different types of data without human entry to back and forth will help reduce costs, unify the sources of data dispersed in a single strategy and strengthen governance by keeping everything that is in accordance with regulations.
Although it is difficult to say exactly how these solutions will take place in 2025, Watkins considers it a story of two halves: many challenges, but also a lot of progress. “Expect a few surprises,” he adds laughing, “but if I could predict them all, I would probably sail on my megayacht at the moment.”